XXII. Priority Research Projects
Many of the most important 2030 Census questions cannot be resolved through advocacy alone. They require legal analysis, empirical testing, independent modeling, and research that can challenge or validate Census assumptions.
The projects below consolidate the most important research needs identified throughout this guide. Some are appropriate for universities or independent statistical researchers. Others may require civil rights organizations, legal scholars, state data centers, philanthropies, or partnerships with communities directly affected by the census design.
The objective should be to produce usable findings before the relevant Census decisions are locked in, not simply retrospective evaluations after 2030.
Project 1: Constitutional Limits of In-Office Enumeration
Commission a focused legal analysis of the circumstances under which Census may count a household through administrative records or other in-office methods without obtaining a direct response.
The analysis should examine the Census Clause, Title 13, Utah v. Evans, and other relevant census cases, with particular attention to the distinction between using administrative information as a limited backstop and relying on it as a substantial alternative to direct enumeration.
Rather than asking whether administrative enumeration is categorically lawful or unlawful, the research should identify the factors most likely to matter legally: the extent of direct contact attempted first, the reliability of the records, the proportion of the population counted this way, the availability of better alternatives, and the risk that the method systematically affects particular populations.
Useful product
A legal memorandum establishing practical guardrails for IOE and identifying design choices that create the greatest litigation risk.
Project 2: Administrative-Record Quality for Historically Undercounted Groups
Conduct an independent assessment of how well the administrative sources likely to support the 2030 Census represent populations historically at greater risk of omission or misplacement.
The work should separately evaluate:
whether a person appears in the records at all;
whether the person is associated with the correct Census Day address;
whether the records reconstruct the household accurately; and
whether individual characteristics are complete and accurate.
Research should pay particular attention to young children, recent movers, immigrants, people experiencing housing instability, tribal populations, complex households, and other groups for whom records may be incomplete or outdated.
The analysis should also distinguish between uses. A dataset may be strong enough to establish that a housing unit is occupied while being much weaker for determining who lives there or describing their characteristics.
Useful product
A population-by-population and variable-by-variable assessment of administrative-record quality that can be compared with Census’s criteria for IOE and the Person Characteristic Frame.
Project 3: Consequences of the Disclosure Avoidance DAO
Model the census products that could realistically be produced under disclosure avoidance methods permitted by the Commerce DAO.
Researchers should apply plausible combinations of suppression, geographic aggregation, category consolidation, rounding, and other permissible techniques to prior census data and reproduce major 2030 use cases.
The analysis should show what happens to:
block-level redistricting data;
tract and neighborhood statistics;
detailed race and ethnicity tables;
tribal and rural geographies; and
small-population cross-tabulations.
The goal is not to advocate for a particular disclosure avoidance method in advance. It is to establish concrete measures of what information would survive each plausible approach and where privacy protection creates unacceptable losses in utility.
Useful product
Public demonstration datasets and a comparison of privacy and utility across realistic non-noise alternatives.
Project 4: Language Access and Census Quality
Estimate the operational consequences of reducing different components of the Census language program.
Research should avoid treating “language assistance” as a single intervention. It should separately examine translated internet questionnaires, telephone response, bilingual mailings, field language support, translated guides, and partner materials.
For each component, researchers should estimate effects on self-response, mode choice, field workload, proxy response, administrative enumeration, data completeness, and cost.
The strongest analysis would identify where language services reduce downstream census costs by increasing direct response, rather than evaluating them solely as an additional expenditure.
Useful product
A mode-by-mode estimate of how multilingual services affect response, cost, and data quality, including the consequences of eliminating or reducing each service.
Project 5: Independence and Accuracy of Coverage Estimation
Evaluate whether the redesigned Coverage Estimation program is sufficiently independent to identify errors in an increasingly administrative-data-driven census.
The central question is whether the census and its quality measure may rely on some of the same incomplete records, matching systems, or assumptions. If the same person is missing from both systems, the apparent agreement between them could conceal rather than reveal an undercount.
Researchers should identify shared data sources and methods, model the potential for correlated error, and evaluate whether targeted independent interviews are sufficient to detect it.
The project should also examine whether additional probability-based field collection is needed among cases where the census and administrative sources appear to agree.
Useful product
An independent assessment of correlated-error risk and recommendations for the minimum amount and design of truly independent coverage measurement.
Project 6: The Future of Census Community Partnerships
Study whether the community infrastructure that supported the 2020 Census can realistically be rebuilt for 2030.
The research should examine more than the number of potential partners. It should assess organizational funding, staffing, trusted-messenger capacity, relationships with Census, and willingness to encourage participation.
Particular attention should be paid to how federal policies affect that willingness. Organizations may have difficulty promoting the census if language access is reduced, important demographic content is removed, or communities fear that government data could be used for immigration enforcement or other harmful purposes.
The project should include direct interviews with organizations that participated in 2020 as well as organizations serving populations likely to require substantial outreach in 2030.
Useful product
A national assessment of partnership capacity and trust, including gaps by geography and population and recommendations for rebuilding the infrastructure before peak outreach begins.
Project 7: Continuous Address Updating Versus Traditional Canvassing
Evaluate whether continuous, data-driven address maintenance produces an address frame that is at least as complete and accurate as the more intensive canvassing approaches used in prior censuses.
The analysis should focus on differential error rather than national averages. It should compare performance in rural and remote areas, tribal communities, colonias and informal settlements, rapidly developing areas, places with hidden or subdivided housing units, and locations where ordinary mailing addresses do not accurately identify residences.
Researchers should examine both omissions and erroneous inclusions and determine whether targeted fieldwork successfully identifies the places where in-office updating performs poorly.
Useful product
An independent comparison of address-frame accuracy by community type, along with criteria for determining where field verification should be mandatory.
Project 8: Near Real-Time Quality Control and Algorithmic Bias
Test whether the automated systems used to detect anomalies, resolve conflicting responses, select primary records, identify duplicates, and direct quality-control interventions perform differently for less common households and populations.
The concern is not limited to conventional demographic bias. Algorithms may perform poorly whenever a household does not resemble the patterns represented most heavily in the training or historical data. That could include multigenerational households, shared-custody families, highly mobile people, unusual address structures, detailed racial and ethnic groups, or households whose names and relationships are represented inconsistently across administrative systems.
Research should examine false matches, missed matches, incorrect duplicate removal, inappropriate anomaly flags, and human overrides. It should also evaluate whether automated quality-control systems are capable of detecting systematic error rather than merely unusual individual cases.
Useful product
A differential-performance audit of the major matching, resolution, and quality-control algorithms, with recommended thresholds for human review and ongoing monitoring.
Building the Research Agenda
These projects should not operate as eight separate research silos. Several depend on the same underlying evidence. Administrative-record quality affects IOE, the PCF, and Coverage Estimation. Address-frame quality affects fieldwork as well as administrative enumeration. Language access and community trust affect whether households ever reach the point where Census must rely on outside records.
Funders and research organizations should therefore coordinate projects, share data and methods where possible, and sequence the work around Census decision points.
The most urgent projects are those capable of influencing Baseline 2, the replacement disclosure avoidance system, and the design entering the 2028 Dress Rehearsal. Other research can continue later in the decade, but findings that arrive after software, contracts, and production rules are fixed will have much less ability to improve the 2030 Census.
The common standard should be practical: each research project should end with findings that can change a decision, establish a guardrail, or give advocates an independent benchmark against which Census claims can be evaluated.